Stop pricing migrations by the terabyte or the hour

I have read a lot of warehouse migration quotes. Almost all of them were priced in one of three units: terabytes, hours or tokens. Not one of those units is the thing that decides what a migration costs.

That mismatch is why migration budgets are the least trusted number in most IT plans, and why sponsors defer the decision for another year. Fix the unit and the number becomes something a CFO can sign.

What actually drives the cost of a warehouse migration?

Priced by terabyte, Estate A costs twenty times as much. In reality Estate B is the harder program by a wide margin. Every migration budget I have seen built on terabytes was wrong. Every one built on object counts was close.

Why do hours-based migration quotes keep moving?

Because the estimate is the work. A systems integrator prices effort, and effort cannot be known until someone has read the procedures, which is most of the job.

That creates incentives nobody intends. When an object turns out harder than expected, the hours go up and so does the invoice. When the team finds a faster way, the hours go down and so does the vendor's revenue. Nobody is acting in bad faith; the unit simply rewards the slow path. The result is a range that widens as the project goes on, with the buyer as the only party who carries the risk and has no way to reduce it.

Are LLM-based migration tools cheaper?

They can look cheaper at the start, and they are open-ended by design. An LLM-based converter charges for what the model reads and writes. Cost scales with every object, every retry and every rerun after a failed test, and nobody can tell you in advance how many reruns a gnarly procedure will need.

There is also no fair baseline to compare against. No LLM-based converter has a public record of completed full-estate migrations with per-object proof attached, so any percentage saving quoted against one is a number with nothing underneath it. We do not quote one. The honest contrast is simpler: a fixed price against a meter.

How the four pricing models compare

How does fixed-price migration work?

There are no hourly, token or usage charges on any rung.

1. Estate assessment, $3,000. Three business days. The full inventory, complexity bands, lineage, dead code, and one fixed price for the whole estate, valid for 90 days.
2. First Domain, $25,000 to $40,000 fixed. One real domain, up to 1,500 objects, converted, reconciled and live in five to six weeks, with a reconciliation threshold in the contract. Miss it and the fee is waived or credited.
3. Full Estate, fixed by size tier. The price from the assessment, with the assessment and First Domain fees deducted, paid 40% at signature, 40% at reconciliation and 20% at go-live.
4. Enterprise license. For estates large enough to run the software themselves, with no caps on objects or runs.

What changes when the price is fixed?

The number is better, and the conversation changes.

The CFO can approve it because it does not move. Procurement can compare it because it is one line, not a rate card. The sponsor can defend it to the board because it came from counting the estate rather than estimating it. And the vendor has every reason to finish quickly, because finishing quickly is the only way its margin improves.

We can offer a fixed price because the conversion is deterministic and the proof is automatic. A method that produces the same output every time can be priced the same way every time.

What a fixed price does not include

Be wary of any fixed price that does not say what is outside it. Ours excludes target-platform costs, the movement of data between platforms (a storage task the reconciliation then checks), re-pointing downstream reports, changes to business logic, and sources outside the published coverage list. If a quote does not tell you what it leaves out, the gaps are where the overruns live.